Signed-off-by: Bertrand Bellenot <bertrand.bellenot@cern.ch>

Signed-off-by: Bertrand Bellenot <bertrand.bellenot@cern.ch>

Signed-off-by: Bertrand Bellenot <bertrand.bellenot@cern.ch>

Signed-off-by: Bertrand Bellenot <bertrand.bellenot@cern.ch>

Now the displayed mean is included between min and max again

now, if the chain did not run yet upon a call to Report, the event loop
is triggered. This is useful when an analysis on cut efficiencies is
to be performed without having to expose the Run method in the interface
and without requiring to book fake actions.

This extensive refactoring standardizes and simplifies the way `TDataFrameActions`
and `Operations` work together.

- logic should be clearer
- avoid the creation of an intermediate shared pointer and lambda function
  that were passed to the `TDFAction`
- `TDFActions` are now templated on the `Operation` type rather than on the
  type of a callable, possibly reducing the total number of template
  specializations of large applications
- `TDFAction` now directly stores the relevant operation as data member
- `TDFAction::Run` calls `Operation::Exec`
- `~TDFaction` calls `Operation::Finalize`
- `ForeachSlot` is now implemented in terms of an `Operation`
  like all other actions

This reverts commit f5ff4110.

This extensive refactoring standardizes and simplifies the way `TDataFrameActions`
and `Operations` work together.

- logic should be clearer
- avoid the creation of an intermediate shared pointer and lambda function
  that were passed to the `TDFAction`
- `TDFActions` are now templated on the `Operation` type rather than on the
  type of a callable, possibly reducing the total number of template
  specializations of large applications
- `TDFAction` now directly stores the relevant operation as data member
- `TDFAction::Run` calls `Operation::Exec`
- `~TDFaction` calls `Operation::Finalize`
- `ForeachSlot` is now implemented in terms of an `Operation`
  like all other actions

This is not only for consistency, but because variadic arguments are greedy
if we want operations with a variadic number of parameters then `slot` should
be first. This change will allow better interplay between TDFActions and
Operations in the (short) future.

- avoid moving callables in AddSlotParameter
- make better use of move contructors
- text fixes

Because the address of the element of a bitset and vector<bool> can not be
taken, we need to use a temporary storage to implement the interface that
requires the address of the element.  So far this had only been implemented
in the specialized collection proxy (TGenBitsetProxy and TGenVectorBoolProxy).
Those are obtained only when cloning a generic collection proxy.
Consequently using the result TClass::GetCollectionProxy for those two
directly did not work in all cases/interfaces.  However since TTree is making
a copy of the collection proxy, it was correctly in all cases.

This patch resolves all the issues seen in ROOT-8574 including
  - the bitset proxy's iterator did not record it started but still recorded its advance
  - the genertic proxy At function in case of bitset or vector<bool> was not return the right address

We also simplified TGenVectorBoolProxy since we know the actual container type at compilation time.

In 2004 kBool_t was the same value as kUChar_t (both 11).
So when the collection proxy were introduced to distinguished the two, a new
one was introduced kBOOL_t (value 21) to be used internally by the collection
proxies.

However this value is returned by TVirtualCollectionProxy::GetType and lead
to incorrect behaviors, for example:

    std::cerr << TDataType::GetDataType(proxy->GetType())->AsString(proxy->At(i)) << "\n";

as the value 21 is in TDataTypes.h the value of kDataTypeAliasUnsigned_t

In 2005 (eab7cae0), kBool_t was updated in TDataTypes.h (to the value 18)
to be distinct from kUChar_t as (at the time) sizeof(bool) on MacOS was 4.
However the code in collection proxy was never modernize to leverage this new
distinction.

Since the value is only recorded in the CollectionProxy objects and those
are never persisted, we can safetly switch from kBOOL_t to kBool_t.

For backwards compatibility

Introduces a new templated WrappedMultiTF1 class and templates all of the the abstract classes it depends on.
Necessary for templating on different vectorization backends and types.

With this we avoid the need to build an std::function from free
functions to pass it as an argument to WrappedMultiTF1Templ